Well, after only two months of use, my water cooled PNY 5090 suffered a fate that seems to have affected a number of 5090s...and it melted.

Sitting there there playing Civilisation V of all things (hardly demanding) I smelt burning and took about 30 seconds to locate it coming from the PC. Shut down and noted that several of the wires on the 12VHWPR Cable had melted. Unfortunately the plastic had also melted into the 5090 socket, rending the card unusable.

This was with a Montech Titan 1200W power supply using all the OEM cables. Not sure what else I could have done to prevent this occuring.

Biggest PITA nowis having to remove the water block, re-pacakge the card in its original cooler, send it back to the retailer and wait for them to "assess" it.

For what its worth, the PC had shown no signs of trouble at all - Had been running wonderfully. The cable was fitted straight and not bent at all. Goes to show, stuff happens!

https://imgur.com/a/EOGl3e7

https://imgur.com/XnJ4Ktp

 

 

 

Edited July 22, 2025Jul 22 by KL Oo

Will be interested to hear what the retailer has to say about the fact you modified the card with an aftermarket liquid cooling block.

There is a number of 5090's already burned, not even including the 4090's suffering the same fate.

Those cables have no load balancing (i mean the electronics behind the drivers). Each 5080 or above is an accident waiting to happen, i thought this was well known.

What this means is that ANY of the tiny cables which are part of the 12VPWR connector can, at any time, transport way more current than its maximum spec, risking a fire which can obviously spread and cause serious problems. Good luck.

Edited July 22, 2025Jul 22 by Nuno Pinto

3 hours ago, Nuno Pinto said:

Those cables have no load balancing (i mean the electronics behind the drivers). Each 5080 or above is an accident waiting to happen, i thought this was well known.

What this means is that ANY of the tiny cables which are part of the 12VPWR connector can, at any time, transport way more current than its maximum spec, risking a fire which can obviously spread and cause serious problems. Good luck.

If I ever did acquire a 5090, I think I'd either modify the power cabling (or have custom cabling built) with 10A inline fuses in each of the 12v main power conductors.  That way no single wire or connector pin can be overloaded to the point of mechanically damaging the cable or connector.

25 minutes ago, Bob Scott said:

If I ever did acquire a 5090, I think I'd either modify the power cabling (or have custom cabling built) with 10A inline fuses in each of the 12v main power conductors.  That way no single wire or connector pin can be overloaded to the point of mechanically damaging the cable or connector.

They already have a slow burn fuse built-in.. to the connector. 😄

They CAN fix it with a proper implementation, but i guess nvidia is not really worried about that with all the income from the AI GPU stuff :(

7 hours ago, somiller said:

Will be interested to hear what the retailer has to say about the fact you modified the card with an aftermarket liquid cooling block.

Not sure the cooling solution can be blamed for what is, evidently, a very widely known design problem with power delivery. But well see....

12 hours ago, KL Oo said:

Not sure what else I could have done to prevent this occuring.

I’ve attached two temp sensor from the motherboard to the either side of the wires and glued the sensors on.  Holding me over until my more tech solution arrives:

https://www.thermal-grizzly.com/en/wireview-gpu-pro/s-tg-wv-90-p-h1n

Rob.

I'm using a 12VHPWR cable with my 5090.  Bought an IR thermometer and under sustained 580W load I'm getting 50C on the cards cable connector, up to 56C on the cable itself, GPUZ shows the GPU die reaching 76C and the thermometer indicates the back of the die on the 5099 reaching 75C.

6 hours ago, Nuno Pinto said:

They CAN fix it with a proper implementation, but i guess nvidia is not really worried about that with all the income from the AI GPU stuff 😞

IIRC the PCIe power spec says that all the power lines are joined. The 3090 (which did have separate power delivery circuits) was technically out of spec.

Cheers

6 hours ago, Luke said:

IIRC the PCIe power spec says that all the power lines are joined. The 3090 (which did have separate power delivery circuits) was technically out of spec.

Cheers

My "knowledge" only comes from information gathered on YouTube by watching quite a few videos from GamersNexus, der8auer and the sorts. Although they have access to way more information than the regular John Doe has, we always have to take this information with a grain of salt.

They did test this problem like i mentioned, but i guess only the ones who build the video cards know for sure what they're doing in there. It does seem feasible that unbalanced current flowing through the connectors can cause these issues, but who knows.

5 hours ago, Nuno Pinto said:

It does seem feasible that unbalanced current flowing through the connectors can cause these issues, but who knows.

No one disagrees that this is the cause. The challenge is that the spec says that all the connectors need to be electrically joined so IIUC (I am not an EE, nor did I date one in college) one cannot measure differential current. The right way to do is keep the connectors separate until you get to the VRM on the card.

My point is that it's not like nVidia did a shoddy implementation, it's a weakness in the underlying spec.

Cheers

24 minutes ago, Luke said:

implementation, it's a weakness in the underlying spec

But nVidia are one of many companies in the consortium that define the PCIe spec.  And nVidia does have a loud voice in this consortium.

nVidia were and are aware of how much load their GPU will/would carry.  They elected to go the cheap route rather come up with a better PCIe solution that they knew would delay to market.

3 hours ago, SayAgain said:

But nVidia are one of many companies in the consortium that define the PCIe spec.  And nVidia does have a loud voice in this consortium.

I'm not sure why you are so insistent on blaming solely nVidia here. As you point out, they are one of many companies and while they are influential they don't have a veto.

Also keep in mind that the PCI-SIG works on specifications far in advance. They released PCIe 7.0 this year, whereas only recently have we been able to buy PCIe 5.0 motherboards. The connector issues have definitely been a problem, but keep in mind that standards bodies work on very different timescales than hardware, never mind software.

Cheers